You are on page 1of 13

Evaluating Risk in Real Estate

By
William C. Wheaton
Raymond G.Torto
Petros Sivitanides
Jon Southard
CB Richard Ellis/TortoWheaton Research
June 1999

The rapid recovery of US real estate markets and the concomitant


emergence of new development have stimulated concern over whether U.S.
real estate markets will once again overbuild as they did in the late 1980s. The
1989-92 crash actually resulted from the confluence of two events; almost six
years of continuous and excessive real estate development mixed together with
a macroeconomic recession. The last time an economic contraction of demand
occurred precisely at the point when real estate markets were saturated was in
1975. On the other hand, in the recession of 1980-82 real estate markets were still
largely under-served, and escaped the economic downturn relatively
unscathed. This review of the historic gyrations in the market reminds us that real
estate risk involves both sides of the market interacting together. The greatest
down side risk comes from a negative economic shock at the apex of a building
boom, while up side risk occurs when positive economic events coincide with
the absence of new supply.
Against this background, the following paper reviews some traditional
definitions and measures of real estate investment risk, and then proposes a
different and more forward-looking methodology for evaluating real estate
markets and property specific risk.

Defining Real Estate Risk: Future Uncertainty


Most economists agree that risk is best conceptualized as greater
uncertainty over likely future outcomes [Bodie, Kane, Markus, 1993]. Technically,
this is expressed as an increase in the (mean preserving) spread of the probability
distribution of future outcomes, whether the outcome is that of a space market
(e.g., rent levels) or an asset market (e.g. return). If the event tends to have a
stationary time series, then this spread is represented by the variance in the
future expected value of the outcome. In more complicated non-stationary
stochastic processes (e.g. Brownian Motion) there is also a variance, in this case
representing the spread in the movement of the outcome. Whatever debate

there is among economists or finance theorists concerns, not so much the


definition of risk, but rather its measurement.

Measuring Real Estate Risk: Historic or Forecast Return Variability


The most often used measures of investment risk have been developed in
traded securities markets, where analysts frequently look at the historic variability
of outcomes (in this case, investment return). This historic variability can be
decomposed into two components: predictable and non-predictable. The
former tends to appear when the series has auto regressive or other distinct
patterns which give the analyst a good idea of at least short-term future
behavior. A series that is quite predictable in one sense has less risk because
the future can be forecast with some degree of confidence. A series that is
largely unpredictable has almost complete uncertainty about its future and in
this sense great risk. Now, traded securities markets are widely held to be
efficient, and as such have little predictable component. If returns are largely a
random walk, then future returns cannot be predicted. Historic variability is both
all the analyst can use, as well as an appropriate measure of likely future
investment risk.
The question is whether real estate risk should be measured as it has been
done in the public markets. The problem with extending such measurements to
real estate is that this category of assets is still largely privately owned, where it is
widely believed that there is less than efficient asset pricing. With inefficient
pricing, positive (uncertain) shocks will always set off a rise and fall in asset prices
that easily can be predicted. Numerous writers have documented the unusually
strong predictable component in real estate markets [Shiller]. When there are
predicable components to real estate returns, these returns can and should be
forecast forward rather than simply using historic variability. Some investment
consultants in fact are using historical return data to predict real estate returns
into the future, using the variability in these returns as a measure of real estate
risk. This is dangerous.
Returning to the discussion on the sources of historic variability, the same
decomposition applies to the future. A forecast of future real estate investment
returns is itself likely to be possibly quite uncertain. The total future risk of real
estate assets has two components: the variability of the forecast return and the
uncertainty surrounding that forecast. The variability of the future forecast by
definition is based only on the predictable component (the forecast model ).
As such, it ignores the random factors that generated much of the historic
variability. Lest there be any doubt about this, look at an econometric forecast
for any variable it is always smoother than its historic data series. Forecasts are
quite useful, but they understate the true likely future variability of the data being
forecast.
In some sense, a real estate forecast that has asset income and price
varying over time because of predictable factors (the model ) actually has less
risk than its predictable variability. With modern financial techniques, it is often

possible to insure hedge or plan for the predicted variability. A much greater
source of true return risk is the uncertainty that the forecast will not be correct.

Measuring Real Estate Risk: Forecast Return Uncertainty


Any forecast of future real estate performance by definition incorporates
only what is predictable and not what is unknown or random. If the forces
influencing the historic movements of an investment are well understood, then
forecast models should fit well and the predictable component is large relative
to what is unknown. In this case there is confidence in the forecast of investment
performance and hence lower risk (uncertainty). Econometrically, the forecast
is said to have tight confidence intervals. Alternatively, if the forecast model has
a low fit and wide confidence intervals, this generates considerably more
uncertainty about the true likely outcome. It is the uncertainty of the forecast
that generates most of the true future risk rather than whether the forecast has a
smooth or varying pattern of returns.
In real estate forecasting there is an additional complication because
such forecasts are almost always contingent . In contingent forecasting, the
underlying model incorporates local market demand and supply dynamics, and
makes assumptions about a range of influential but exogenous
macroeconomic variables (economic growth, interest rates). Most real estate
forecasters (Torto Wheaton Research, FW Dodge, Property Portfolio Research,
Rosen Consulting) base their forecasts on some VAR or vector autoregressive
system 1(either with or without structural restrictions). These in turn, however, are
always conditional on a forecast of national and regional economic variables
(provided by vendors such as RFA, DRI or WEFA).
These contingent real estate forecasts have two sources of uncertainty:
uncertainty in what is assumed or forecast about the macroeconomic variables,
and the uncertainty or fit of the forecast model conditional on these variables.
Both sources of uncertainty can be important and can interact with each other
in complicated ways. Modern time series analysis has largely given up trying to
derive equations for the confidence intervals of a multi-variable VAR type model
even when it is unconditional. Most often, numerical bootstrapping simulations
are used to develop such confidence intervals [Hamilton]. Similarly, if model error
is ignored, simulation approaches can be used to produce forecast confidence
intervals for the real estate market if the confidence intervals are known for the
macroeconomic inputs. It is far too complicated, however, to either derive or
simulate the joint likelihood of both the macroeconomic inputs and the VAR
model error.
The forecasting models we have developed (which predict rents,
construction and property income) tend to fit statistically quite well given a
regional and macroeconomic outlook. Looking retrospectively over our
1

Vector autoregressive models relate a system of variables to their own lagged values,
but not to current values.

forecasts, much of the forecast error has been due to unforeseen


macroeconomic changes and unpredictable regional economic or
demographic factors. For example, in the mid 1980s we predicted a national
real estate crunch at the end of the decade, but not with the severity that
actually occurred. The reason is that our forecasts (at that time) did not
incorporate a macroeconomic recession. Similarly, while recoveries in many
rocky mountain markets were forecast for the late 1980s and early 1990s, the
regional employment outlooks we used did not foresee the emerging economic
strength of Denver and Salt Lake. Thus, much of the uncertainty or risk in our real
estate forecasts is due not to real estate model error, but to unforeseen events or
shocks likely to happen to the local economy together with how each
property market reacts to these shocks.
We have devised a system for ascertaining the risk of real estate investing
in each market and property type that incorporates these two concerns. It is
important to recognize that this system does not encompass all of the risk that
real estate is exposed to, but it does cover the main risk that market
fundamentals will deviate from that which is anticipated. It operates in these
three steps:
1). First, we begin by developing confidence intervals for the regional
economic variables that underlie the forecasts that is the likelihood of unknown
local economic events happening. We do this by estimating an ARIMA forecast
for the employment data that drives each local office market. Our statistical
software package contains a feature that generates statistical confidence
intervals (plus and minus 1 and 2 standard deviations) around the forecast trend
in each variable.
2). Since the ARIMA trend cannot consider a range of future
macroeconomic factors that will influence each areas economy (e.g., the baby
boom, changing industrial structure), we take these confidence bands and
apply them to a vendors regional forecast that does (RFA).
3). Finally, we use our models to forecast the real estate market under
each economic confidence band, and apply the scenario probabilities to the
corresponding real estate forecast. This generates confidence intervals for rent,
construction, etc.
This approach gives results that make considerable sense. Local
economies that have strong and relatively smooth historic growth trends (e.g.,
Atlanta, Phoenix) tend to have generally tight economic confidence bands.
Areas with slower long term growth and greater cyclic oscillations (e.g.,
Cleveland, Boston) have wider economic confidence bands. When the
economic forecasts are fed into the real estate model, each areas supply
elasticity and current stage in the building boom considerably influence the
width or tightness of the rent and construction confidence bands that emerge.
This can best be described by examining the following cases.

Los Angeles: High Economic, Low Supply Risk


The economy of Los Angeles has a number of risks associated with its
future growth. First, its traditional economic base (entertainment, aerospace) has
always reacted strongly to macro-economic events. Second, its economic
structure is changing rapidly. Applying our ARIMA generated confidence bands
to a forecast for the Los Angeles economy generated by RFA, we find the results
in Figure 1. These show the base forecast together with the one and two
standard deviation confidence intervals in that forecast. The 2 standard
deviation scenario, for example, suggests that there is only a 2.5% probability
that ten years from now, Los Angeles would have the same (or fewer) number of
office workers as today. At the other extreme, there is also a 2.5% probability that
by 2008, office employment would have grown by more than 60% (from 880 to
1,340 workers). In between are office employment outcomes which are far more
likely.
Each of these economic outlooks is fed into our office forecasting model
for the Los Angeles market. The model forecasts rents and construction
simultaneously as each interacts with and influences the other. The outcomes
are shown Figures 2 and 3. The 2 standard deviation economic scenario leads
to a cessation of construction (from its current rather low level) and to flat rents
over the next five years (in current dollars). After 2004, rents begin to rise with
inflation. A scenario this poor happens with only a 2.5% probability. In the +2
standard deviation case, construction rises to 8 million square feet annually in
2004, while rents experience a 60% increase between 1998 and 2008. Again, an
outcome at least this optimistic occurs with only a 2.5% probability. In between
are rent and construction projections that are again far more likely.
In some sense, the office market income risk in these figures looks quite
low. The spread in year 2008 maximum/minimum rent levels is 41% of todays
value and even in the worst case, gross income would be flat. All other
outcomes have income rising significantly. The reasons for this are two-fold. First,
the LA office market is just beginning a new development cycle. This is because
rents have only recovered back to the levels of the late 1980s in current dollars.
They are still below their historic highs in real dollars, and it is the latter which is a
much better benchmark for the cost of replacement. A negative shock at this
point does little damage to the market. Positive shocks will generate a few years
of strong rental growth before new supply catches up. Secondly, office supply in
LA historically is somewhat rent inelastic . It takes significant increases and
some time to generate a development boom in reaction to rental growth. It
has been argued that this makes real estate markets in general more stable
[Wheaton] and therefore less likely to overbuild.

Figure 1: Office Employment, Los Angeles


1400

Off. Emp. (x1000)

1300
1200
1100

-2 S D
-1 S D
B ase Case
+1 S D
+2 S D

1000
900
800
700
1988

1993

1998

2003

2008

Year

Figure 2: Rent Index, Los Angeles


35

Rent Index ($/SF)

30

25

-2 S D
- 1S D
B ase Case
+1 S D

20

+2 S D

15

10
1988

1993

1998

2003

2008

Ye ar

Figure 3: Completions, Los Angeles


16,000

Completions (SFx100)

14,000
12,000
10,000
-2S D
- 1S D
B ase Case
+1S D
+2 S D

8,000
6,000
4,000
2,000
0
1988

1993

1998
Year

2003

2008

Washington D.C.: Low Economic, High Supply Risk


Applying the same ARIMA methodology to the Washington DC economy,
together with the RFA forecast for that region, yields the office employment
confidence intervals shown in Figure 4. Washington has less economic risk than
Los Angeles. The range of 2008 employment levels between the +2 and -2
standard deviation scenarios is 44% of current office employment, as opposed to
57% for Los Angeles. The economic forecast is less risky because the Federal
presence in Washington historically has smoothed its economy and because
growing high technology employment seems to be filling the void left by
shrinking defense related sectors.
In Washington, however, the next office construction cycle is already well
underway. Completions this year were 4.5 million, and 10 million is under
construction for 1999. Beyond that, committed planned projects loom even
bigger. Turning to the rent forecasts in Figure 5, the reason for this building boom
becomes apparent. Washington rents are already 30% ahead of their late 1980s
levels. Even adjusting for inflation, real rents in the Washington market have set a
new precedent. Historically, the building cycle in Washington has been quite
volatile and exhibits a great deal of momentum once underway. The market has
a high rental elasticity of construction, and is prone to overbuilding, with a
boom well under way.
The rental outlook for Washington shows considerable risk. In the 2
standard deviation scenario, rents in 2008 fall back to 1992 levels, while in the +2
standard deviation outlook, they are only 20% higher than currently. While
outcomes above or below these bands are unlikely (2.5% probability), the
forecasts in between show little rental growth and considerable oscillation over
the next decade. The range in year 2008 rent levels is 61% of todays value as
opposed to only 40% for Los Angeles. The culprit is again the already occurring
building boom. The alternative economic scenarios only determine the extent
and magnitude of new development too much momentum is already
underway to avoid overbuilding. 2

While the confidence intervals of the economic forecast are symmetric around their
expected values, the property market reactions need not be. The ability to generate
almost any level of new construction tends to limit rental values on the upside, while the
durability of built assets makes rent more volatile on the downside.

Figure 4: Office Employment, Washington, D.C.


1200

Off. Emp. (x1000)

1100
1000
900
-2 S D
-1 S D
B ase Case
+1 S D
+2 S D

800
700
600
500
1988

1993

1998

2003

2008

Year

Figure 5: Rent Index, Washington, D.C.

Rent Index ($/SF)

35

30

25
-2 S D
- 1S D
B ase Case

20

+1 S D
+2 S D

15

10
1988

1993

1998

2003

2008

Year

Figure 6: Completions, Washington, D.C.


16,000

Completions (SFx1000)

14,000
12,000
10,000
-2S D
- 1S D
B ase Case

8,000
6,000

+1S D
+2 S D

4,000
2,000
0
1988

1993

1998
Year

2003

2008

Translating Market Risk into Property Income Risk


The confidence bands for the Washington and Los Angeles office markets
are academically interesting, but to be truly useful for investment decision
making, they must be translated into the likely scenarios for a particular
propertys income. In many instances, the longer term and staggered leasing of
a property can provide some insurance against temporary market rent
fluctuations. At the same time, the magnitude and growth of operating
expenses creates a leverage effect on gross income increasing the volatility of
NOI or net cash income. To compare each areas market risk for a
representative property we must convert each market rent forecast into one
for net property income. We have devised a software that does this in several
steps with user supplied inputs.
1). Office rents are assumed to rollover every five years and move with the
forecast market rent. Thus, gross income is a five year blend of the rent series
depicted. An allowance is made for lease non-renewals, and the downtime
during re-leasing is based on the market wide vacancy rate at that period.
2). Prototypical operating costs have been obtained from several sources,
and are assumed to rise with CPI inflation. Additional prototypical allowances
are made for tenant improvements after lease non-renewals, and capital
expenses. Using these, prototypical building gross income is converted to net
operating income after capital expenses (here called NOI).
In Figures 7 and 8, this approach is applied to a prototypical 100,000
square foot office building in each market. In the case of Los Angeles, there is
only a small risk that building net income will fall below $400,000 from its initial
level of roughly $500,000. This risk is greatest in the year 2008, but is still only about
2.5%. In Los Angeles the expected income stream of a property seems exposed
to very little downside risk. The implications for lending in this market are quite
important. For example, a typical 70-80% debt service coverage ratio would be
quite secure and virtually risk free. Washington, however, presents another story.
The initial property income of $750,000 stands a 50% probability of falling below
$600,000 in the year 2008 and has a 15% probability of dropping to below
$200,000 in that year. Thus, a loan with the same 70-80% debt service coverage
ratio in Washington has almost a 50% probability of being delinquent in the year
2008. In fact, for virtually any size loan, there is at least a 2.5% probability of falling
delinquent in 2008 for properties in the Washington DC market.

This type of analysis can be duplicated across property types and


geographic markets and then used to comparatively evaluate lending risk. We
are in the process of undertaking such an exercise, and initial indications suggest
that there are far greater differences in lending risk across markets than current
differences in mortgage spreads would suggest.

Figure 7: N OI, Los Angeles


1,400,000
1,200,000

Dollars ($)

1,000,000
800,000

-2 S D
-1 S D

600,000

T W B as e
+1 S D

400,000

+2 S D

200,000
0
-200,000
1997

1998

1999

2000

2001

2002

2003

2004

2005

2006

2007

Ye ar

Figure 8: N OI, W ashington, D .C .


1,400,000
1,200,000

Do llars ($)

1,000,000
-2 S D

800,000

-1S D

600,000

T W B ase
+1S D

400,000

+2 S D

200,000
0
-200,000
1997

1998

1999

2000

2001

2002

2003

2004

2005

2006

2007

Y e ar

Summarizing Property Income Risk: IRR Risk


The two examined MSAs show substantial difference in their degrees of
market risk, and resulting income risk, all as measured by the spread in these
respective forecasts. The issue remains as to how these forecast spreads can be
captured into a single measure of total return that could be used to rank or
compare investments. An obvious candidate should calculate the IRR that is
implied by the income streams such as those shown in Figures 7 and 8.
To calculate the IRR for a prototypical property of course requires an
initial property value. We have gathered several sources of current building
transaction data and used these to derive a cap rate for assets in each market.

In LA this value is 8.15%, while in Washington it is 8.90%. This cap rate is applied to
the base year NOI numbers to calculate an estimated transaction price for
the prototypical building asset. The income forecast is then run for a long term
period (55 years) effectively assuming a perpetual hold. Then, calculate the IRR
that equates this property value with the perpetual net income forecast.
If this process is repeated for each economic scenario, then the
probability distribution of IRRs can be obtained. In Los Angeles, this approach
yields the five IRR estimates shown in the first column of Table 1 below. Assuming
a normal distribution of the scenarios, we can calculate the expected value and
standard deviation of these income returns in Los Angeles (and then
3
Washington). In the table, Washington not only has a lower expected IRR, but
also higher variance as well. Since individual real estate assets, or even specific
geographic portfolios of these assets are not traded in any current public
market, there is little reason to expect that they will follow CAPM rules.
Earlier, we observed that the Washington D.C. economy is less risky than
that of Los Angeles. Despite this lower economic risk, the real estate investment
risk in Washington D.C., as judged by the standard deviation in IRR, is somewhat
greater than in Los Angeles. The reason, of course, is the greater volatility of
supply and that a building boom is well underway in Washington. As an
experiment, we have generated a series of forecasts for Washington that use the
economic confidence bands of the Los Angeles economy (around the base RFA
Washington forecast). The results are shown in the final column of Table 1. With
this equivalent (and greater) amount of economic risk, the standard deviation in
Washingtons IRR rises to 1.78. Thus, the greater supply problems in Washington
are responsible for about 40 basis points of relative extra risk in comparison to
Los Angeles.

Table 1: IRR with 55 year hold


Scenario Los Angeles Washington, D.C. Washington, D.C. (LA)
11.01
7.34
6.36
-2 SD
12.49
10.12
9.76
-1 SD
13.99
11.89
11.89
TW Base
15.36
12.98
13.12
+1 SD
16.76
13.76
14.07
+2 SD
13.94
11.49
11.38
Mean IRR
1.92
2.23
3.17
Var IRR
1.39
1.49
1.78
StDev IRR

Assuming the IRR distribution is normal, the five observations available can quite closely
approximate the distributions mean and standard deviation.

In examining Table 1, we are struck by two overall observations regarding


our approach. First, our expected IRR values are roughly comparable with
historic NCREIF data and seem quite in line with the current investment targets of
many asset managers. Second, the risk measures produced by our analysis are
far lower than the historic risk in total return that emerges from the NREIF data.

Conclusions: Sources of Investment Risk


We have outlined an approach for assessing the income risk in properties.
The approach is based on the premise that the most important source of this risk
is the markets fundamentals rent and vacancy and that this risk derives from
uncertainty about demand (the markets economy) and supply (stage of
building cycle). Using modern time series analysis, it is possible to quantify this
kind of risk. We suggest that the confidence intervals around a future forecast of
each markets economy can be fed into a real estate model to generate
confidence intervals for prototypical property rents and income. With
confidence bands around property income, we can calculate an expected
(perpetual hold) IRR and the standard deviation of that return.
If this approach yields risk measures that seem low by historic performance
standards, it is because it accounts for only one source of investment risk: income
risk. In examining the NCREIF series, for example, much of the historic variability in
returns has come from capital gains and losses. While these appraisal based
valuations have been often criticized, it is certainly true that the approach
described here does not consider pricing risk, and with anything less than a
perpetual hold, this type of risk can be important.
While the income risk measures here would make it difficult to compare
real estate risk with that of other (more traded) assets, it would seem possible to
make comparisons within the real estate sector, across markets and types of
property. In fact, using the type of expected IRRs and standard deviations shown
in Table 1, we have begun to undertake a more formal portfolio analysis
exercise. With a set of mean and standard deviation IRRs across 54 markets and
four property types, we have found that an efficient real estate frontier
involves a relatively small number of area-property types, and that several
different portfolios can come quite close to the efficient one in terms of risk
adjusted return.

References
Bodie, Z, A. Kane, and A. Markus, Investments, Irwin (Boston), 1993
Hamilton, John, Modern Times Series Analysis, Wiley and Sons (New York), 1996.
Shiller, R, and K. Case, The efficiency of the market for Single Family Homes ,
AER, 77, 3 (1989) pp111-222.
Wheaton, W. Real Estate Cycles: Some Fundamentals , Real Estate Economics,
22,3 (summer) 1999 (forthcoming).

You might also like